The Automated Vehicle and Electric Vehicle Sentiment Indices separately track the positive and negative sentiments
expressed on
Twitter regarding these two subjects.
The indices are updated daily based on a sample of up to 5,000 tweets per
day on each subject. The tweets are selected based on a custome set of
keywords. Retweets, replies, and quote tweets were not included in the data
set, only original tweets. In addition, the tweets are filtered, so that
identical or nearly identical tweets (a very common occurrence due to the
number of automated bots on Twitter) are filtered out.
The resulting set of tweets are then analyzed using
VADER, a
popular open source tool that was developed and optimized for conducting
sentiment analysis on social media posts such as tweets. Some minor
additions tailored to the subject matter were made to the lexicon that is
included in VADER. From each daily sample, two measures are computed. The
Score, or average sentiment, is the simple average of all sampled tweets,
where negative tweets are assigned a value of -1, neutral tweets a 0, and
positive tweets a +1. The score, therefore, ranges between -1 (every tweet
negative) to +1 (ever tweet positive). The positive/negative ratio throws
out the neutral tweets and is simply the ratio of positive to negative
tweets. The complete ratio, using all three values, is displayed as a donut
graph.
In addition, the 100 most frequently occurring one and 2-word n-grams are
determined for each set of tweets. These are displayed as word clouds. Using
the word cloud data, a sample of the words that appear in the current day's
cloud, but not in the cloud from seven days ago is listed under "What's Hot"
while a sample of those that were in the word cloud 7 days ago, but not in
the current cloud are listed under "What's Not."
There have been a number of studies using sentiment analysis to examine
attitudes concerning automated and electric vehicles, including several that
looked at changes in sentiment caused by crashes. However these have been
limited to either a single snapshot or a short term comparison across a
couple of weeks. In addition, those studies tailored the search terms for a
specific crash event, rather then measuring from a constant baseline across
long time spans and multiple incidents. A key goal of this project is to
provide daily tracking of sentiment over an extended period. This allows
trends to be monitored as well as measuing the effect that events such as
automated vehicle crasehes have on the sentiment scores.and how quickly they
recover. The first two months of data collection coincided with a
significant incident involving a Tesla in self-driving mode, and the AV
sentiment data showed a large, significant increase in the number of
negative tweets concerning automated vehicles for about a week afterwards.
More information can be found in a short write up:
Observing the Effect of a Crash on Twitter
Sentiment: Early Results from Time Series Data.